Social Science Statistics Blog

This reminds me a lot of an introductory essay that Richard McKelvey wrote for a special feature in the Proceedings of the National Academy of Sciences. The title of the essay is "The Hard Sciences". The first two paragraphs are:

In at least one respect, the social, economic, political, and behavioral sciences truly are the "hard" sciences. A problem that is unique to these areas of research is that the subjects of the study (human beings) can read. Because of this, developing a theory to understand and predict an election outcome or a stock market crash is fundamentally more difficult than the problem of predicting a chemical reaction or an earthquake.

In the case of the chemical reaction or the earthquake, the publication of the theory will not have any effect on the prediction. In the case of the election outcome or the stock market crash, if the prediction is public, and the theory is convincing, individuals may have incentives to take advantage of their knowledge of the theory to alter their behavior. Thus, if individuals believe a prediction that a stock market crash is imminent, they may decide to sell off declining stocks before the crash, implying that the crash will occur earlier than predicted. So the original prediction will be wrong. Any good theory of behavior in the social sciences must work even when the participants know the theory, that is, the theory must survive its own publication. This characteristic of a theory is captured mathematically by concepts of game theoretic equilibrium such as the Nash equilibrium [the general existence of which was first established in these pages 49 years ago] for noncooperative games and the core for cooperative games. The necessity of a theory to be publication-proof helps explain why game theoretic models have become central to the study of social science.

The full cite is PNAS, Vol 96. p. 10549. September 1999.

That sounds like a very interesting article (and a good point). Sort of the Heisenburg Uncertainty Principle of the social sciences: the best theories are those in which the process of theorising doesn't disrupt the thing being theorised about.

On the other hand - and just to quibble a bit with the word "best" - I imagine there is also scope for theories that either (a) are about domains that nobody will read the theory or know about it, e.g., economies in, say, a preliterate society; or (b) contain within the theory predictions about how behavior would change when people know about the theory. I don't know enough about economics to know how often either of these sorts of theories occur, but they seem possible in principle, no?